One of the best ways to find signal in the noise of a major technological shift is through the time-tested journalistic tactic of following the money. Back in January I predicted that in B2B markets, AI Can’t Cost This Much. I argued that “pricing the future based on the cost of the present is a losing bet … However many trillions will be spent on AI over the course of the next decade, one thing will remain constant: There are huge opportunities to be exploited in finding more efficient and less expensive ways to deliver technology to both consumers and businesses.”
Nearly three years before writing that post, I made another prediction, this time in consumer markets: It would soon be normal for consumers to pay as much as $200 a month for AI services. Combined, we pay more than that for mobile phone, cable, and internet services, after all.
Have both of my predictions come true? While it’s certainly not “normal,” millions of us are in fact paying $100 a month (or more) for ChatGPT Pro, Claude Max, and/or Gemini. And tens of millions of us are paying at least $20 for basic AI subscriptions. It’s absolutely in the interests of these AI providers to upsell us – think of the cost of your very first cable bill compared with the last one you paid. I quit cable a few years ago, but before I did, my bill averaged more than $200 a month.
So what about this year’s prediction, that AI costs will come down significantly in the enterprise/business world? The answer to this one feels far more certain. Here are three headlines from my morning reading today:
Microsoft Replaces OpenAI, Anthropic With Own AI in Some Apps (Bloomberg)
AI Giants Are Handing Out Tons of Free Computing Power to Grab Startup Share (WSJ)
China’s Answer to AI Sticker Shock (The Atlantic)
Taken together, these three stories paint a picture of a market acting rationally in the face of downward pricing pressure. Microsoft is cutting its OpEx by replacing expensive frontier models with cheaper versions it built in-house. Those frontier models are attempting to secure replacements for that lost revenue by luring startups into using their services through steep discounts and introductory pricing offers. Meanwhile, Chinese models built with far less expensive inputs (DeepSeek, Z.ai) are starting to take over the enterprise market – recent data suggests nearly 40 percent of US corporate AI use now depends on these far cheaper models.
The tech industry spent the lion’s share of the past year figuring out how to pay for investments in data centers and chip foundries with fixed cost assumptions that may no longer pencil out. With the AI marketplace in the process of a significant realignment, it certainly feels like there’s a fair bit of economic headwind ahead. There’s also plenty of opportunity – if we let the free market do its work.
—